import hipdnn
import torch


def build_resample_graph(
    hipdnn_handle,
    torch_tensor_x,
    scale,
    resample_mode,
    coordinate_transform_mode,
    generate_index,
    hipdnn_data_type,
):
    # Create graph
    graph = hipdnn.pygraph(
        handle=hipdnn_handle,
        io_data_type=hipdnn_data_type,
        intermediate_data_type=hipdnn.data_type.FLOAT,
        compute_data_type=hipdnn.data_type.FLOAT,
        name="resample",
    )

    # Create hipdnn tensors
    hipdnn_tensor_x = graph.tensor_like(torch_tensor_x)

    # Create resample op
    hipdnn_tensor_y, _ = graph.resample(
        x=hipdnn_tensor_x,
        scale=scale,
        resample_mode=resample_mode,
        coordinate_transform_mode=coordinate_transform_mode,
        generate_index=generate_index,
        name="resample",
    )
    hipdnn_tensor_y.set_output(True)
    graph.build(hipdnn_handle)

    return (graph, hipdnn_tensor_x, hipdnn_tensor_y)


if __name__ == "__main__":
    # Input dimensions
    n = 4  # Batch size
    c = 16  # Number of input channels
    h = 8  # Height
    w = 8  # Width
    scale = [3, 3]
    resample_mode = hipdnn.resample_mode.BILINEAR
    coordinate_transform_mode = hipdnn.coordinate_transform_mode.COORDINATE_ASYMMETRIC
    generate_index = False

    hipdnn_data_type = hipdnn.data_type.HALF
    torch_data_type = torch.float16

    torch_tensor_x = torch.rand(n, c, h, w, dtype=torch_data_type, device="cuda").to(
        memory_format=torch.channels_last
    )

    hipdnn_handle = hipdnn.create_handle()

    graph, hipdnn_tensor_x, hipdnn_tensor_y = build_resample_graph(
        hipdnn_handle,
        torch_tensor_x,
        scale,
        resample_mode,
        coordinate_transform_mode,
        generate_index,
        hipdnn_data_type,
    )

    torch_tensor_y = torch.empty(hipdnn_tensor_y.get_dim(), dtype=torch_data_type, device="cuda")
    variant_pack = {
        hipdnn_tensor_x: torch_tensor_x.data_ptr(),
        hipdnn_tensor_y: torch_tensor_y.data_ptr(),
    }
    workspace = torch.empty(graph.get_workspace_size(), dtype=torch.uint8, device="cuda")

    graph.exec(variant_pack=variant_pack, workspace=workspace.data_ptr())
    print("Resample graph execution complete.")